An End-to-End System for Crowdsourced 3d Maps for Autonomous Vehicles: The Mapping Component

TL;DR

This paper introduces a cost-effective, end-to-end system for crowdsourcing high-definition 3D maps for autonomous vehicles, achieving sub-20cm accuracy using consumer-grade sensors and real-time processing.

Contribution

It presents the first end-to-end HD mapping pipeline in global coordinates using only consumer-grade sensors and real-time triangulation for autonomous vehicle applications.

Findings

Achieved less than 20cm mean absolute accuracy at sign corners.

Utilized real-time triangulation and offline clustering for map construction.

Demonstrated feasibility with only consumer-grade sensors in real-world journeys.

Abstract

Autonomous vehicles rely on precise high definition (HD) 3d maps for navigation. This paper presents the mapping component of an end-to-end system for crowdsourcing precise 3d maps with semantically meaningful landmarks such as traffic signs (6 dof pose, shape and size) and traffic lanes (3d splines). The system uses consumer grade parts, and in particular, relies on a single front facing camera and a consumer grade GPS. Using real-time sign and lane triangulation on-device in the vehicle, with offline sign/lane clustering across multiple journeys and offline Bundle Adjustment across multiple journeys in the backend, we construct maps with mean absolute accuracy at sign corners of less than 20 cm from 25 journeys. To the best of our knowledge, this is the first end-to-end HD mapping pipeline in global coordinates in the automotive context using cost effective sensors.